OpenRouter's latest data reveals that Chinese AI large language models have surpassed their US counterparts in usage for the fifth consecutive week. For the week of March 30 to April 5, 2026, global AI model consumption reached 27 trillion tokens, up 18.9% week-over-week. Chinese models accounted for 12.96 trillion tokens, a gain of 31.48%, while US models managed only 3.03 trillion tokens, a meager 0.76% increase.
Chinese Models Sweep Top Six Spots
Notably, all six most-used AI models globally this week originated from China. Alibaba's Qwen3.6 series dominates: Qwen3.6 Plus (free tier) leads with 4.6 trillion tokens, followed by Qwen3.6 Plus Preview at 1.64 trillion tokens (third place). The remaining four models also stem from Chinese firms, underscoring the nation's growing influence in open-source LLM ecosystems.
OpenRouter Data Highlights Compute Shift
OpenRouter, an AI model aggregation platform, reflects real developer and user choices. China's sustained lead is not a one-off spike but a five-week structural trend. Analysts attribute this to aggressive open-source strategies and lower usage costs, combined with rapid domestic AI application deployment in areas like customer service, content generation, and code assistance.
US Growth Stalls, Concentration Weakens
US model growth slowed to just 0.76% this week. Excluding marginal increases from OpenAI and Anthropic, many smaller US models experienced token declines. Industry observers worry that export controls and compute restrictions may be eroding the US developer ecosystem. While US models still excel in complex reasoning tasks, their high cost limits mass adoption.
Implications for Crypto and Blockchain
The AI compute boom is reshaping cryptocurrency markets. AI-themed tokens (e.g., Fetch.ai, SingularityNET) have attracted capital amid the compute narrative, while decentralized compute networks (RNDR, Akash) could benefit from a global rebalancing of compute resources driven by Chinese models. The latest data reinforces the investment thesis for AI-blockchain convergence, though short-term speculative risks remain.